Huat Thart Chiang

The series alignment algorithm allows for the quantitative comparison between molecular simulations and experimental characterization.

A retrieval-augmented language model framework is presented for generating real-space structures and simulating their small-angle scattering (SAS) profiles in Python. In this workflow, the language model translates user requests into executable code that constructs the desired geometry, which is then passed to the validated Monte Carlo Distribution Function Method (MC-DFM), a numerical solution to the Debye scattering equation that computes the scattering intensity. Domain-specific instructions and examples are supplied through retrieval to improve reliability and consistency. The approach is demonstrated on a range of benchmark and user-defined systems, including spheres, cylinders, core-shell particles, spherical clusters, anisotropic cube assemblies, superlattices, protein-based cargo-cage structures, and tube-like protein assemblies. The framework captures expected scattering features for monodisperse, polydisperse, and mixed assemblies, including contrast variation and invariant-based weighting for complex superlattice populations. These results show that retrieval-augmented language models can serve as practical interfaces for rapid and reproducible small-angle scattering model generation and data analysis.

The rapid emergence of agentic AI presents new opportunities and challenges for accelerating scientific discovery through tool-augmented reasoning, autonomous workflows, and reproducible results. To explore these capabilities in a hands-on, community-driven setting, we hosted the Agentic AI for Science Hackathon 2025, which attracted 352 registered participants. The event was designed to benchmark, test, and extend agentic AI workflows across scientific tasks while remaining modelagnostic and lowering the barrier to entry for participants from diverse backgrounds. Participants engaged with a unified API ecosystem centered on the fully open-access AtomGPT.org API (AGAPI), implementing tool calling, asynchronous agents, and multi-model reasoning to solve problems spanning materials database retrieval, literature search, mathematical reasoning, and scientific workflow automation. The hackathon prompt emphasized reproducibility, transparent tool usage, and explicit agent-tool interaction, encouraging participants to move beyond single-prompt question answering toward structured, auditable pipelines. Outcomes included working agentic prototypes, identification of failure modes in contemporary chatbots, and actionable insights into best practices for agent design in scientific contexts. This paper documents the hackathon design, task structure, and key lessons learned, highlighting how hackathon-based evaluations can complement formal benchmarks and inform the development of robust agentic AI systems for science. Notebooks from participants who consented to public sharing are provided in the supplementary information.

DNA molecules, modified with cholesterol functional groups, are physically anchored to lipid-encapsulated silica nanoparticles to drive their colloidal assembly as the process is analyzed with scattering and simulations.

Single-stranded DNA molecules modified with cholesterol functional groups are physically tethered to silica nanoparticles (Diameter 25 nm) that are encapsulated in a lipid bilayer. This enhances the lateral mobility of the DNA molecules across the nanoparticle surface and enables nonspecific bonding, eliminating the need for specialized surface chemistries (such as silane or thiol ligands). To induce assembly, double-stranded DNA ‘bridge’ molecules are then added with complementary nucleotides to the DNA ‘anchor’ molecules that are physically tethered to the lipids on the surface of the particles. Assembly is observed to occur at room temperature and without the need for temperature annealing. The lateral mobility of DNA on f luid-supported nanoparticles introduces an added degree of freedom that enables the particles to adjust their positions relative to one another, which could be advantageous for the assembly process. Using automated liquid handling tools, assemblies are created in high throughput and rapidly characterized using SAXS. It is determined that the relative concentration of DNA-to-silica and the ionic strength of the solution are important parameters that affect the resulting assembly. Analysis of SAXS data is performed using coarse-grained particle dynamics simulations as a function of the interparticle potential. The results support the spontaneous formation of semi-crystalline particle assemblies by particle condensation, where the interparticle distance is tuned by the sequence of the DNA ‘bridge’ used to link the particles. Crystallinity analysis performed on the resulting simulations, optimized to match SAXS observations, suggest that particle clusters display increased crystallinity in the center of the clusters, but their maximum size remains relatively small (100’s of nm) before settling occurs, which limits the extent of crystallization.

Directional interactions that generate regular coordination geometries are a powerful means of guiding molecular and colloidal self-assembly, but implementing such high-level interactions with proteins remains challenging due to their complex shapes and intricate interface properties. Here we describe a modular approach to protein nanomaterial design inspired by the rich chemical diversity that can be generated from the small number of atomic valencies. We design protein building blocks using deep learning-based generative tools, incorporating regular coordination geometries and tailorable bonding interactions that enable the assembly of diverse closed and open architectures guided by simple geometric principles. Experimental characterization confirms the successful formation of more than 20 multicomponent polyhedral protein cages, two-dimensional arrays and three-dimensional protein lattices, with a high (10%–50%) success rate and electron microscopy data closely matching the corresponding design models. Due to modularity, individual building blocks can assemble with different partners to generate distinct regular assemblies, resulting in an economy of parts and enabling the construction of reconfigurable networks for designer nanomaterials.

Structure elicitation from small angle scattering (i.e. SAXS, SANS) curves of large biomolecular assemblies is notoriously challenging. This is because the simulation of high-resolution features in the structure of large macromolecular assemblies, such as de-novo protein assemblies, is computationally demanding when it needs to cover a broad range of length scales. Conventional methods such as the numerical approximation to the Debye equation or the use of spherical harmonics do not scale well as the size of the assembly increases, which limits their application to small structures (e.g. individual proteins). This work explores the effectiveness of a Monte Carlo method to simulate and fit scattering curves for large biomolecular assemblies spanning over ranges covering atomic and molecular detail, as well as large scale (100’s of nm) features. Due to its excellent speed and scalability, it can be combined with a fitting algorithm to extract structural features from experimental small-angle scattering curves in biomolecular assemblies that are otherwise intractable for interpretation. This work first demonstrates the effectiveness of the tool using experimental small angle X-ray scattering (SAXS) data from tile-like proteins that assemble into 1D tube-like macromolecular structures. The diameter distribution of tubes is extracted from SAXS fits, and this quantitatively compared with distributions from electron microscopy. SAXS data is also obtained from 2D sheet-like protein assemblies and the proposed method is used to quantify structural features such as the separation distance between protein building blocks and the flexing of the sheet. An open-source implementation of the methodology is provided for use in a broad range of biological systems involving multi-scale scattering analysis.

Autonomous experimentation, or self-driving labs as they are popularly known, is a novel way of experimental planning and scheduling to systematically integrate automated material synthesis, characterization, and data analysis for accelerated materials design and discovery. This paper presents an autonomous experimentation workflow to conduct an iterative, on-demand synthesis, and structural characterization of colloidal gold nanoparticles to map composition information to nano-scale structure in a closed loop. We introduce a unified experimental planning framework based on differentiable models of the shape of a spectrum to solve the two categories of problems tackled using autonomous experimentation: a) quantitatively mapping compositional parameters to regions with particular behavior in structure or property i.e. phase mapping; b) inverse design of materials with a target structure or function i.e. material retrosynthesis. Using functional data analysis, we describe a data-driven computational model to map correlations between the compositions (i.e., synthesis or processing conditions) to the shape of a characterization curve (e.g.: UV-Vis spectroscopy). By integrating generative pre-training, active learning, and high-throughput experimentation, we train a surrogate model capable of predicting characterization curves across compositional spaces. The model-based approach to experimental planning provides convergence criteria for terminating iterative experimentation by measuring the model's test accuracy. We demonstrate the utility of trained differentiable models in generating phase maps of a seed-mediated growth of colloidal gold nanoparticles for extracting design rules, secondary interaction effects (between synthesis components and their impact on nanoparticle structure), and efficient navigation of different gold nanoparticle morphologies space. We then describe an application of the differentiable models to colloidal gold nanoparticle retrosynthesis using a gradient-based optimization to showcase the construction of a self-driving lab doubly capable of structure optimization and phase map generation.

Circular dichroism (CD) spectroscopy has emerged as a potent tool for probing chiral small-molecule ligand exchange on natively achiral quantum dots (QDs). In this study, we report a novel approach to identifying QD-biomolecule interactions by inducing chirality in CdS QDs using thermoresponsive elastin-like polypeptides (ELPs) engineered with C-terminal cysteine residues. Our method is based on a versatile two-step ligand exchange process starting from monodisperse oleate-capped QDs in nonpolar media and proceeding through an easily accessed achiral glycine-capped QD intermediate. Successful conjugation of the ELPs onto the QDs is confirmed by the diagnostic CD response corresponding to the QD electronic transitions in the visible range. The resulting ELP:CdS conjugates demonstrate thermally reversible coacervation, as observed through dynamic light scattering, small-angle X-ray scattering, and electron microscopy. This research provides a foundation for using induced chirality in QD electronic transitions to probe QD conjugation to complex peptides and proteins, opening pathways for designing dynamic, stimuli-responsive hybrid nanomaterials.

We present an autonomous data-driven framework that iteratively explores the experimental design space of silver nanoparticle synthesis to obtain control over the formation of a desired morphology and size. The objective of the method is to identify design rules such as the effects of the design variables on the structure of the nanoparticle. The framework balances multimodal characterization methods (i.e. UV-vis spectroscopy, SAXS, TEM), taking into account the cost of performing a measurement and the quality of information gained. By integrating with an AI agent, we identify important design variables in the synthesis of small colloidally stable plate-like silver particles and outline how each variable affects plate thickness, radius, polydispersity, and relative concentration. Our findings are consistent with the literature, demonstrating that the framework could be further applied to new systems that have not been well characterized and understood. The framework is generalizable and allows tangible knowledge extraction from the high-throughput experimental runs while still considering inherent stochasticity.